I've had the "pleasure"of testing out how well uTorrent, Azurues, and KTorrent work when they are unable to talk to the tracker. The results are pretty expected (at least for me).

The Torrents:
Torrent "A": New-release TV series episode, ~5000 seeds, 3000 peers
Torrent "B": Enthusiast Linux distro, 25 seeds, 1 peer.

Both had roughly a 40/40/10 makeup of Azureus, uTorrent, and Other respectively.

Testing Methodology
First, a Python script using BitTorrent 5.0.0's libraries was used to strip announce URL's and tracker lists from the torrents. Tracker lists were removed while the announce URL was replaced with a nonexistent one.

All torrents were allowed to run for approx. 5 minutes, at which time the # of connected peers/seeds and download rate were measured.

The Clients:
Azureus 2.5.0.0, Ubuntu Linux 6.10, Sun Java 1.6.0
KTorrent 2.1 from SVN, Ubuntu 6.10, latest snapshot updated before test.
uTorrent 1.6.0.0, Windows XP

All clients were configured with no upload or download caps on a LAN with wondershaper. 500KB/s down / 50KB/s up bandwidth, no limit to # of connections.

The results:

Torrent "A":
KTorrent: 195 connections [DHT claims finding 300 peers], 450KB/s down / 40KB/s up
uTorrent: 210 connections [client claims finding 3000-some peers], 450KB/s down, 40KB/s up
Azureus: 180 connections [DDB claims finding 200 peers, verbose logging indicates via PEX it has a list of the entire swarm]

Torrent "B":
KTorrent: 5 connections [DHT claims 10 peers], 10 down, 0 up, leech was BitTornado and thus not found.
uTorrent: 23 connections [client claims finding all 25 peers], 50KB/s down, 10KB/s up
Azureus: 24 connections [DDB found 3 peers, verbose logging indicates rest were found via PEX from a located Azureus client], 60KB/s down, 8KB/s up



Conclusions:
On a large torrent, one can located enough peers from DHT alone that a decent speed can be maintained. PEX doesn't really matter all that much in this case.

On a smaller torrent, DHT alone can usually find enough peers to get started, but having PEX is a great reward.

On the large TV torrent, I will note the announce list had 8 trackers, so it's fairly incredulous that all would go down. On the small torrent there was only one tracker, which could very well go down. In fact, I did remember one such tracker going down due to Digg effect.



* Measurements are not made scientifically, not meant as a comparative performance analysis between these three clients, more to demonstrate the usefulness of DHT/PEX in a situation where trackers couldn't be contacted.

Interesting results. I wonder why KT as problems connecting to peers on torrent B. AZ and uT had around 25, compared to 5 in KT - I would expect a bit more...

So to conclude, you want PEX

Post Posted: 28 Dec 2006 11:20 pm Post subject:
Interesting results. I wonder why KT as problems connecting to peers on torrent B. AZ and uT had around 25, compared to 5 in KT - I would expect a bit more...

Well, not everybody runs DHT, even if the client supports it. And then of course AZ has their own incompatible version of DHT, this rules out getting into contact with AZ peers, unless they find you via PEX.

Ivan wrote:Interesting results. I wonder why KT as problems connecting to peers on torrent B. AZ and uT had around 25, compared to 5 in KT - I would expect a bit more...

Well, basically PEX. With DHT, KTorrent will be able to locate a good chunk of uTorrents, same with uTorrent. However, as soon as uTorrent connects to another uTorrent, it gets a full list of every member of the swarm courtesy of PEX